Added L2 cache size property to CUDA backends

Author: fjwillemsen

kernel_tuner/backends/cupy.py

@@ -46,6 +46,7 @@ def __init__(self, device=0, iterations=7, compiler_options=None, observers=None
         self.devprops = dev.attributes
         self.cc = dev.compute_capability
         self.max_threads = self.devprops["MaxThreadsPerBlock"]
+        self.cache_size_L2 = self.devprops["L2CacheSize"]
 
         self.iterations = iterations
         self.current_module = None

kernel_tuner/backends/nvcuda.py

@@ -66,6 +66,10 @@ def __init__(self, device=0, iterations=7, compiler_options=None, observers=None
             cudart.cudaDeviceAttr.cudaDevAttrMaxThreadsPerBlock, device
         )
         cuda_error_check(err)
+        err, self.cache_size_L2 = cudart.cudaDeviceGetAttribute(
+            cudart.cudaDeviceAttr.cudaDevAttrL2CacheSize, device
+        )
+        cuda_error_check(err)
         self.cc = f"{major}{minor}"
         self.iterations = iterations
         self.current_module = None

kernel_tuner/backends/pycuda.py

@@ -101,6 +101,7 @@ def _finish_up():
             str(k): v for (k, v) in self.context.get_device().get_attributes().items()
         }
         self.max_threads = devprops["MAX_THREADS_PER_BLOCK"]
+        self.cache_size_L2 = devprops["L2_CACHE_SIZE"]
         cc = str(devprops.get("COMPUTE_CAPABILITY_MAJOR", "0")) + str(
             devprops.get("COMPUTE_CAPABILITY_MINOR", "0")
         )